One Brain. Four Hands.
Give your AI coding assistant permanent memory, honest code insight, and security awareness — all local, all open, all yours.
What Forge is becoming.
Forge is the open, local-first memory and intelligence layer for AI-assisted development. Your AI assistant — Claude Code, Cursor, Windsurf, your own scripts — knows your codebase structure, your security posture, your past mistakes, and your decisions. Not because you typed it twice. Because everything writes to one Brain you own, on your machine.
Four tools. One brain.
Every tool writes to the same Brain. Your AI reads from all of them.
Six releases. One direction.
Each ships when its gate is met — not on a calendar.
12 repos, a working 4-agent harness, sutra on npm.
Give your AI a permanent memory in one command.
Honest code-graph and session insight — free, local, 60-sec wow.
Your memory, everywhere. Team brains. Free hosted tier, generous rate limits.
Your AI now knows your repo — structure, risk, history.
A collective brain for AI devs. Plug in, share, learn.
What ships in each release.
Brain stops being a substrate and becomes a product a stranger can adopt in 5 minutes.
- 1-command install — generator at mcp.sbknext.com
- Memory survives /clear and /compact
- Auto-capture decisions and mistakes (no manual save)
- Session-start recall injects repo context
- Web dashboard — browse, search, edit memories
- Free tier — instant token, no card
Free, shareable tools that bring devs in — each ends with "now give it memory → Brain."
- Live watch mode — graph updates as you code
- Cross-repo map (echo-ai ↔ brain-api)
- Shareable graph link
- "Explain this feature" — AI explains from the graph
- Real-time mistake alerts mid-session
- Skill auto-templating from repeated edits
- Token + cost analytics per session
- "Promote to ledger" — one click feeds Brain
Retention and adoption. Forge is free; open-core (Apache-2.0) is yours to self-host. Need scale or compliance? Self-host or contact us.
- Multi-device sync
- Team / shared brains
- SSO and audit log (enterprise self-host)
- Free hosted tier — generous rate limits, no paywall
- Import notes, ledgers, git history
- CI gate — fail PRs on new secrets vs baseline
- One-line GitHub Action
- Trend-over-time dashboard
- Deeper Python + multi-language depth
The convergence. All tools enrich one Brain. The AI gets smarter for it.
- Unified knowledge write from sutra, secure, fuse
- Codebase-aware AI queries (no re-scan)
- Memory consolidation — summaries decay gracefully
- Decision log linked to commits
- Personal Mistakes Ledger, productized
From personal brain to collective intelligence.
- Plugin API public — extend Brain without forking
- Shared knowledge graphs — opt-in community brains
- forge-skills marketplace
- Cross-tool, cross-team memory federation
- "Forge ships Forge" harness opened as a template
12-month swimlane.
Approximate cadence — gates drive real ship dates.
Every repo. A destination.
Where each piece of the ecosystem is headed.
brain-mcp / corebrainThe product — Brain (Cloud + open-core)forge-sutraFunnel #1 — free, live-watch, feeds Brainclaude-fuseFunnel #2 — auto-capture engineforge-secureSecond tool — CI gate (month 3+), free + self-hostableforge-clientThe SDK every tool installsforge-siteOne-story front door → Brainforge-skills3.0 marketplace seedforge framework / harnessThe story — #1 content engine, free and openecho-aiProof-of-Forge demoDream features. Not committed.
Ideas we want to build — no promises, no dates.
Natural-language Q&A over sutra graph + Brain.
Flags orphaned endpoints + dangling tests on every PR.
Feature graph + memory recall inside the editor.
secure finds a secret, opens a PR to remove + rotate.
Harness auto-digest to your build-in-public feed.
See what you knew about this repo a month ago.
Point Forge at a repo, get a new-hire walkthrough.
Full end-to-end with Ollama/llama/qwen — zero cloud.
No date-driven shipping. Each release ships when its gate is met. Forge is free — generous rate limits, no paywall. Need to scale, run private, or meet compliance/SSO? Self-host the open-core (Apache-2.0), or contact us for a one-time setup. We build in public — follow along.